dm-raid1.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 2003 Sistina Software Limited.
   4 * Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
   5 *
   6 * This file is released under the GPL.
   7 */
   8
   9#include "dm-bio-record.h"
  10
  11#include <linux/init.h>
  12#include <linux/mempool.h>
  13#include <linux/module.h>
  14#include <linux/pagemap.h>
  15#include <linux/slab.h>
  16#include <linux/workqueue.h>
  17#include <linux/device-mapper.h>
  18#include <linux/dm-io.h>
  19#include <linux/dm-dirty-log.h>
  20#include <linux/dm-kcopyd.h>
  21#include <linux/dm-region-hash.h>
  22
  23static struct workqueue_struct *dm_raid1_wq;
  24
  25#define DM_MSG_PREFIX "raid1"
  26
  27#define MAX_RECOVERY 1	/* Maximum number of regions recovered in parallel. */
  28
  29#define MAX_NR_MIRRORS	(DM_KCOPYD_MAX_REGIONS + 1)
  30
  31#define DM_RAID1_HANDLE_ERRORS	0x01
  32#define DM_RAID1_KEEP_LOG	0x02
  33#define errors_handled(p)	((p)->features & DM_RAID1_HANDLE_ERRORS)
  34#define keep_log(p)		((p)->features & DM_RAID1_KEEP_LOG)
  35
  36static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped);
  37
  38/*
  39 *---------------------------------------------------------------
  40 * Mirror set structures.
  41 *---------------------------------------------------------------
  42 */
  43enum dm_raid1_error {
  44	DM_RAID1_WRITE_ERROR,
  45	DM_RAID1_FLUSH_ERROR,
  46	DM_RAID1_SYNC_ERROR,
  47	DM_RAID1_READ_ERROR
  48};
  49
  50struct mirror {
  51	struct mirror_set *ms;
  52	atomic_t error_count;
  53	unsigned long error_type;
  54	struct dm_dev *dev;
  55	sector_t offset;
  56};
  57
  58struct mirror_set {
  59	struct dm_target *ti;
  60	struct list_head list;
  61
  62	uint64_t features;
  63
  64	spinlock_t lock;	/* protects the lists */
  65	struct bio_list reads;
  66	struct bio_list writes;
  67	struct bio_list failures;
  68	struct bio_list holds;	/* bios are waiting until suspend */
  69
  70	struct dm_region_hash *rh;
  71	struct dm_kcopyd_client *kcopyd_client;
  72	struct dm_io_client *io_client;
  73
  74	/* recovery */
  75	region_t nr_regions;
  76	int in_sync;
  77	int log_failure;
  78	int leg_failure;
  79	atomic_t suspend;
  80
  81	atomic_t default_mirror;	/* Default mirror */
  82
  83	struct workqueue_struct *kmirrord_wq;
  84	struct work_struct kmirrord_work;
  85	struct timer_list timer;
  86	unsigned long timer_pending;
  87
  88	struct work_struct trigger_event;
  89
  90	unsigned int nr_mirrors;
  91	struct mirror mirror[];
  92};
  93
  94DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(raid1_resync_throttle,
  95		"A percentage of time allocated for raid resynchronization");
  96
  97static void wakeup_mirrord(void *context)
  98{
  99	struct mirror_set *ms = context;
 100
 101	queue_work(ms->kmirrord_wq, &ms->kmirrord_work);
 102}
 103
 104static void delayed_wake_fn(struct timer_list *t)
 105{
 106	struct mirror_set *ms = timer_container_of(ms, t, timer);
 107
 108	clear_bit(0, &ms->timer_pending);
 109	wakeup_mirrord(ms);
 110}
 111
 112static void delayed_wake(struct mirror_set *ms)
 113{
 114	if (test_and_set_bit(0, &ms->timer_pending))
 115		return;
 116
 117	ms->timer.expires = jiffies + HZ / 5;
 118	add_timer(&ms->timer);
 119}
 120
 121static void wakeup_all_recovery_waiters(void *context)
 122{
 123	wake_up_all(&_kmirrord_recovery_stopped);
 124}
 125
 126static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
 127{
 128	unsigned long flags;
 129	int should_wake = 0;
 130	struct bio_list *bl;
 131
 132	bl = (rw == WRITE) ? &ms->writes : &ms->reads;
 133	spin_lock_irqsave(&ms->lock, flags);
 134	should_wake = !(bl->head);
 135	bio_list_add(bl, bio);
 136	if (should_wake)
 137		wakeup_mirrord(ms);
 138	spin_unlock_irqrestore(&ms->lock, flags);
 139}
 140
 141static void dispatch_bios(void *context, struct bio_list *bio_list)
 142{
 143	struct mirror_set *ms = context;
 144	struct bio *bio;
 145
 146	while ((bio = bio_list_pop(bio_list)))
 147		queue_bio(ms, bio, WRITE);
 148}
 149
 150struct dm_raid1_bio_record {
 151	struct mirror *m;
 152	/* if details->bi_bdev == NULL, details were not saved */
 153	struct dm_bio_details details;
 154	region_t write_region;
 155};
 156
 157/*
 158 * Every mirror should look like this one.
 159 */
 160#define DEFAULT_MIRROR 0
 161
 162/*
 163 * This is yucky.  We squirrel the mirror struct away inside
 164 * bi_next for read/write buffers.  This is safe since the bh
 165 * doesn't get submitted to the lower levels of block layer.
 166 */
 167static struct mirror *bio_get_m(struct bio *bio)
 168{
 169	return (struct mirror *) bio->bi_next;
 170}
 171
 172static void bio_set_m(struct bio *bio, struct mirror *m)
 173{
 174	bio->bi_next = (struct bio *) m;
 175}
 176
 177static struct mirror *get_default_mirror(struct mirror_set *ms)
 178{
 179	return &ms->mirror[atomic_read(&ms->default_mirror)];
 180}
 181
 182static void set_default_mirror(struct mirror *m)
 183{
 184	struct mirror_set *ms = m->ms;
 185	struct mirror *m0 = &(ms->mirror[0]);
 186
 187	atomic_set(&ms->default_mirror, m - m0);
 188}
 189
 190static struct mirror *get_valid_mirror(struct mirror_set *ms)
 191{
 192	struct mirror *m;
 193
 194	for (m = ms->mirror; m < ms->mirror + ms->nr_mirrors; m++)
 195		if (!atomic_read(&m->error_count))
 196			return m;
 197
 198	return NULL;
 199}
 200
 201/* fail_mirror
 202 * @m: mirror device to fail
 203 * @error_type: one of the enum's, DM_RAID1_*_ERROR
 204 *
 205 * If errors are being handled, record the type of
 206 * error encountered for this device.  If this type
 207 * of error has already been recorded, we can return;
 208 * otherwise, we must signal userspace by triggering
 209 * an event.  Additionally, if the device is the
 210 * primary device, we must choose a new primary, but
 211 * only if the mirror is in-sync.
 212 *
 213 * This function must not block.
 214 */
 215static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type)
 216{
 217	struct mirror_set *ms = m->ms;
 218	struct mirror *new;
 219
 220	ms->leg_failure = 1;
 221
 222	/*
 223	 * error_count is used for nothing more than a
 224	 * simple way to tell if a device has encountered
 225	 * errors.
 226	 */
 227	atomic_inc(&m->error_count);
 228
 229	if (test_and_set_bit(error_type, &m->error_type))
 230		return;
 231
 232	if (!errors_handled(ms))
 233		return;
 234
 235	if (m != get_default_mirror(ms))
 236		goto out;
 237
 238	if (!ms->in_sync && !keep_log(ms)) {
 239		/*
 240		 * Better to issue requests to same failing device
 241		 * than to risk returning corrupt data.
 242		 */
 243		DMERR("Primary mirror (%s) failed while out-of-sync: Reads may fail.",
 244		      m->dev->name);
 245		goto out;
 246	}
 247
 248	new = get_valid_mirror(ms);
 249	if (new)
 250		set_default_mirror(new);
 251	else
 252		DMWARN("All sides of mirror have failed.");
 253
 254out:
 255	queue_work(dm_raid1_wq, &ms->trigger_event);
 256}
 257
 258static int mirror_flush(struct dm_target *ti)
 259{
 260	struct mirror_set *ms = ti->private;
 261	unsigned long error_bits;
 262
 263	unsigned int i;
 264	struct dm_io_region io[MAX_NR_MIRRORS];
 265	struct mirror *m;
 266	struct dm_io_request io_req = {
 267		.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC,
 268		.mem.type = DM_IO_KMEM,
 269		.mem.ptr.addr = NULL,
 270		.client = ms->io_client,
 271	};
 272
 273	for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) {
 274		io[i].bdev = m->dev->bdev;
 275		io[i].sector = 0;
 276		io[i].count = 0;
 277	}
 278
 279	error_bits = -1;
 280	dm_io(&io_req, ms->nr_mirrors, io, &error_bits, IOPRIO_DEFAULT);
 281	if (unlikely(error_bits != 0)) {
 282		for (i = 0; i < ms->nr_mirrors; i++)
 283			if (test_bit(i, &error_bits))
 284				fail_mirror(ms->mirror + i,
 285					    DM_RAID1_FLUSH_ERROR);
 286		return -EIO;
 287	}
 288
 289	return 0;
 290}
 291
 292/*
 293 *---------------------------------------------------------------
 294 * Recovery.
 295 *
 296 * When a mirror is first activated we may find that some regions
 297 * are in the no-sync state.  We have to recover these by
 298 * recopying from the default mirror to all the others.
 299 *---------------------------------------------------------------
 300 */
 301static void recovery_complete(int read_err, unsigned long write_err,
 302			      void *context)
 303{
 304	struct dm_region *reg = context;
 305	struct mirror_set *ms = dm_rh_region_context(reg);
 306	int m, bit = 0;
 307
 308	if (read_err) {
 309		/* Read error means the failure of default mirror. */
 310		DMERR_LIMIT("Unable to read primary mirror during recovery");
 311		fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR);
 312	}
 313
 314	if (write_err) {
 315		DMERR_LIMIT("Write error during recovery (error = 0x%lx)",
 316			    write_err);
 317		/*
 318		 * Bits correspond to devices (excluding default mirror).
 319		 * The default mirror cannot change during recovery.
 320		 */
 321		for (m = 0; m < ms->nr_mirrors; m++) {
 322			if (&ms->mirror[m] == get_default_mirror(ms))
 323				continue;
 324			if (test_bit(bit, &write_err))
 325				fail_mirror(ms->mirror + m,
 326					    DM_RAID1_SYNC_ERROR);
 327			bit++;
 328		}
 329	}
 330
 331	dm_rh_recovery_end(reg, !(read_err || write_err));
 332}
 333
 334static void recover(struct mirror_set *ms, struct dm_region *reg)
 335{
 336	unsigned int i;
 337	struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest;
 338	struct mirror *m;
 339	unsigned long flags = 0;
 340	region_t key = dm_rh_get_region_key(reg);
 341	sector_t region_size = dm_rh_get_region_size(ms->rh);
 342
 343	/* fill in the source */
 344	m = get_default_mirror(ms);
 345	from.bdev = m->dev->bdev;
 346	from.sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
 347	if (key == (ms->nr_regions - 1)) {
 348		/*
 349		 * The final region may be smaller than
 350		 * region_size.
 351		 */
 352		from.count = ms->ti->len & (region_size - 1);
 353		if (!from.count)
 354			from.count = region_size;
 355	} else
 356		from.count = region_size;
 357
 358	/* fill in the destinations */
 359	for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
 360		if (&ms->mirror[i] == get_default_mirror(ms))
 361			continue;
 362
 363		m = ms->mirror + i;
 364		dest->bdev = m->dev->bdev;
 365		dest->sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
 366		dest->count = from.count;
 367		dest++;
 368	}
 369
 370	/* hand to kcopyd */
 371	if (!errors_handled(ms))
 372		flags |= BIT(DM_KCOPYD_IGNORE_ERROR);
 373
 374	dm_kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to,
 375		       flags, recovery_complete, reg);
 376}
 377
 378static void reset_ms_flags(struct mirror_set *ms)
 379{
 380	unsigned int m;
 381
 382	ms->leg_failure = 0;
 383	for (m = 0; m < ms->nr_mirrors; m++) {
 384		atomic_set(&(ms->mirror[m].error_count), 0);
 385		ms->mirror[m].error_type = 0;
 386	}
 387}
 388
 389static void do_recovery(struct mirror_set *ms)
 390{
 391	struct dm_region *reg;
 392	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
 393
 394	/*
 395	 * Start quiescing some regions.
 396	 */
 397	dm_rh_recovery_prepare(ms->rh);
 398
 399	/*
 400	 * Copy any already quiesced regions.
 401	 */
 402	while ((reg = dm_rh_recovery_start(ms->rh)))
 403		recover(ms, reg);
 404
 405	/*
 406	 * Update the in sync flag.
 407	 */
 408	if (!ms->in_sync &&
 409	    (log->type->get_sync_count(log) == ms->nr_regions)) {
 410		/* the sync is complete */
 411		dm_table_event(ms->ti->table);
 412		ms->in_sync = 1;
 413		reset_ms_flags(ms);
 414	}
 415}
 416
 417/*
 418 *---------------------------------------------------------------
 419 * Reads
 420 *---------------------------------------------------------------
 421 */
 422static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
 423{
 424	struct mirror *m = get_default_mirror(ms);
 425
 426	do {
 427		if (likely(!atomic_read(&m->error_count)))
 428			return m;
 429
 430		if (m-- == ms->mirror)
 431			m += ms->nr_mirrors;
 432	} while (m != get_default_mirror(ms));
 433
 434	return NULL;
 435}
 436
 437static int default_ok(struct mirror *m)
 438{
 439	struct mirror *default_mirror = get_default_mirror(m->ms);
 440
 441	return !atomic_read(&default_mirror->error_count);
 442}
 443
 444static int mirror_available(struct mirror_set *ms, struct bio *bio)
 445{
 446	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
 447	region_t region = dm_rh_bio_to_region(ms->rh, bio);
 448
 449	if (log->type->in_sync(log, region, 0))
 450		return choose_mirror(ms,  bio->bi_iter.bi_sector) ? 1 : 0;
 451
 452	return 0;
 453}
 454
 455/*
 456 * remap a buffer to a particular mirror.
 457 */
 458static sector_t map_sector(struct mirror *m, struct bio *bio)
 459{
 460	if (unlikely(!bio->bi_iter.bi_size))
 461		return 0;
 462	return m->offset + dm_target_offset(m->ms->ti, bio->bi_iter.bi_sector);
 463}
 464
 465static void map_bio(struct mirror *m, struct bio *bio)
 466{
 467	bio_set_dev(bio, m->dev->bdev);
 468	bio->bi_iter.bi_sector = map_sector(m, bio);
 469}
 470
 471static void map_region(struct dm_io_region *io, struct mirror *m,
 472		       struct bio *bio)
 473{
 474	io->bdev = m->dev->bdev;
 475	io->sector = map_sector(m, bio);
 476	io->count = bio_sectors(bio);
 477}
 478
 479static void hold_bio(struct mirror_set *ms, struct bio *bio)
 480{
 481	/*
 482	 * Lock is required to avoid race condition during suspend
 483	 * process.
 484	 */
 485	spin_lock_irq(&ms->lock);
 486
 487	if (atomic_read(&ms->suspend)) {
 488		spin_unlock_irq(&ms->lock);
 489
 490		/*
 491		 * If device is suspended, complete the bio.
 492		 */
 493		if (dm_noflush_suspending(ms->ti))
 494			bio->bi_status = BLK_STS_DM_REQUEUE;
 495		else
 496			bio->bi_status = BLK_STS_IOERR;
 497
 498		bio_endio(bio);
 499		return;
 500	}
 501
 502	/*
 503	 * Hold bio until the suspend is complete.
 504	 */
 505	bio_list_add(&ms->holds, bio);
 506	spin_unlock_irq(&ms->lock);
 507}
 508
 509/*
 510 *---------------------------------------------------------------
 511 * Reads
 512 *---------------------------------------------------------------
 513 */
 514static void read_callback(unsigned long error, void *context)
 515{
 516	struct bio *bio = context;
 517	struct mirror *m;
 518
 519	m = bio_get_m(bio);
 520	bio_set_m(bio, NULL);
 521
 522	if (likely(!error)) {
 523		bio_endio(bio);
 524		return;
 525	}
 526
 527	fail_mirror(m, DM_RAID1_READ_ERROR);
 528
 529	if (likely(default_ok(m)) || mirror_available(m->ms, bio)) {
 530		DMWARN_LIMIT("Read failure on mirror device %s. Trying alternative device.",
 531			     m->dev->name);
 532		queue_bio(m->ms, bio, bio_data_dir(bio));
 533		return;
 534	}
 535
 536	DMERR_LIMIT("Read failure on mirror device %s.  Failing I/O.",
 537		    m->dev->name);
 538	bio_io_error(bio);
 539}
 540
 541/* Asynchronous read. */
 542static void read_async_bio(struct mirror *m, struct bio *bio)
 543{
 544	struct dm_io_region io;
 545	struct dm_io_request io_req = {
 546		.bi_opf = REQ_OP_READ,
 547		.mem.type = DM_IO_BIO,
 548		.mem.ptr.bio = bio,
 549		.notify.fn = read_callback,
 550		.notify.context = bio,
 551		.client = m->ms->io_client,
 552	};
 553
 554	map_region(&io, m, bio);
 555	bio_set_m(bio, m);
 556	BUG_ON(dm_io(&io_req, 1, &io, NULL, IOPRIO_DEFAULT));
 557}
 558
 559static inline int region_in_sync(struct mirror_set *ms, region_t region,
 560				 int may_block)
 561{
 562	int state = dm_rh_get_state(ms->rh, region, may_block);
 563	return state == DM_RH_CLEAN || state == DM_RH_DIRTY;
 564}
 565
 566static void do_reads(struct mirror_set *ms, struct bio_list *reads)
 567{
 568	region_t region;
 569	struct bio *bio;
 570	struct mirror *m;
 571
 572	while ((bio = bio_list_pop(reads))) {
 573		region = dm_rh_bio_to_region(ms->rh, bio);
 574		m = get_default_mirror(ms);
 575
 576		/*
 577		 * We can only read balance if the region is in sync.
 578		 */
 579		if (likely(region_in_sync(ms, region, 1)))
 580			m = choose_mirror(ms, bio->bi_iter.bi_sector);
 581		else if (m && atomic_read(&m->error_count))
 582			m = NULL;
 583
 584		if (likely(m))
 585			read_async_bio(m, bio);
 586		else
 587			bio_io_error(bio);
 588	}
 589}
 590
 591/*
 592 *---------------------------------------------------------------------
 593 * Writes.
 594 *
 595 * We do different things with the write io depending on the
 596 * state of the region that it's in:
 597 *
 598 * SYNC:	increment pending, use kcopyd to write to *all* mirrors
 599 * RECOVERING:	delay the io until recovery completes
 600 * NOSYNC:	increment pending, just write to the default mirror
 601 *---------------------------------------------------------------------
 602 */
 603static void write_callback(unsigned long error, void *context)
 604{
 605	unsigned int i;
 606	struct bio *bio = context;
 607	struct mirror_set *ms;
 608	int should_wake = 0;
 609	unsigned long flags;
 610
 611	ms = bio_get_m(bio)->ms;
 612	bio_set_m(bio, NULL);
 613
 614	/*
 615	 * NOTE: We don't decrement the pending count here,
 616	 * instead it is done by the targets endio function.
 617	 * This way we handle both writes to SYNC and NOSYNC
 618	 * regions with the same code.
 619	 */
 620	if (likely(!error)) {
 621		bio_endio(bio);
 622		return;
 623	}
 624
 625	/*
 626	 * If the bio is discard, return an error, but do not
 627	 * degrade the array.
 628	 */
 629	if (bio_op(bio) == REQ_OP_DISCARD) {
 630		bio->bi_status = BLK_STS_NOTSUPP;
 631		bio_endio(bio);
 632		return;
 633	}
 634
 635	for (i = 0; i < ms->nr_mirrors; i++)
 636		if (test_bit(i, &error))
 637			fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR);
 638
 639	/*
 640	 * Need to raise event.  Since raising
 641	 * events can block, we need to do it in
 642	 * the main thread.
 643	 */
 644	spin_lock_irqsave(&ms->lock, flags);
 645	if (!ms->failures.head)
 646		should_wake = 1;
 647	bio_list_add(&ms->failures, bio);
 648	if (should_wake)
 649		wakeup_mirrord(ms);
 650	spin_unlock_irqrestore(&ms->lock, flags);
 651}
 652
 653static void do_write(struct mirror_set *ms, struct bio *bio)
 654{
 655	unsigned int i;
 656	struct dm_io_region io[MAX_NR_MIRRORS], *dest = io;
 657	struct mirror *m;
 658	blk_opf_t op_flags = bio->bi_opf & (REQ_FUA | REQ_PREFLUSH | REQ_ATOMIC);
 659	struct dm_io_request io_req = {
 660		.bi_opf = REQ_OP_WRITE | op_flags,
 661		.mem.type = DM_IO_BIO,
 662		.mem.ptr.bio = bio,
 663		.notify.fn = write_callback,
 664		.notify.context = bio,
 665		.client = ms->io_client,
 666	};
 667
 668	if (bio_op(bio) == REQ_OP_DISCARD) {
 669		io_req.bi_opf = REQ_OP_DISCARD | op_flags;
 670		io_req.mem.type = DM_IO_KMEM;
 671		io_req.mem.ptr.addr = NULL;
 672	}
 673
 674	for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++)
 675		map_region(dest++, m, bio);
 676
 677	/*
 678	 * Use default mirror because we only need it to retrieve the reference
 679	 * to the mirror set in write_callback().
 680	 */
 681	bio_set_m(bio, get_default_mirror(ms));
 682
 683	BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL, IOPRIO_DEFAULT));
 684}
 685
 686static void do_writes(struct mirror_set *ms, struct bio_list *writes)
 687{
 688	int state;
 689	struct bio *bio;
 690	struct bio_list sync, nosync, recover, *this_list = NULL;
 691	struct bio_list requeue;
 692	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
 693	region_t region;
 694
 695	if (!writes->head)
 696		return;
 697
 698	/*
 699	 * Classify each write.
 700	 */
 701	bio_list_init(&sync);
 702	bio_list_init(&nosync);
 703	bio_list_init(&recover);
 704	bio_list_init(&requeue);
 705
 706	while ((bio = bio_list_pop(writes))) {
 707		if ((bio->bi_opf & REQ_PREFLUSH) ||
 708		    (bio_op(bio) == REQ_OP_DISCARD)) {
 709			bio_list_add(&sync, bio);
 710			continue;
 711		}
 712
 713		region = dm_rh_bio_to_region(ms->rh, bio);
 714
 715		if (log->type->is_remote_recovering &&
 716		    log->type->is_remote_recovering(log, region)) {
 717			bio_list_add(&requeue, bio);
 718			continue;
 719		}
 720
 721		state = dm_rh_get_state(ms->rh, region, 1);
 722		switch (state) {
 723		case DM_RH_CLEAN:
 724		case DM_RH_DIRTY:
 725			this_list = &sync;
 726			break;
 727
 728		case DM_RH_NOSYNC:
 729			this_list = &nosync;
 730			break;
 731
 732		case DM_RH_RECOVERING:
 733			this_list = &recover;
 734			break;
 735		}
 736
 737		bio_list_add(this_list, bio);
 738	}
 739
 740	/*
 741	 * Add bios that are delayed due to remote recovery
 742	 * back on to the write queue
 743	 */
 744	if (unlikely(requeue.head)) {
 745		spin_lock_irq(&ms->lock);
 746		bio_list_merge(&ms->writes, &requeue);
 747		spin_unlock_irq(&ms->lock);
 748		delayed_wake(ms);
 749	}
 750
 751	/*
 752	 * Increment the pending counts for any regions that will
 753	 * be written to (writes to recover regions are going to
 754	 * be delayed).
 755	 */
 756	dm_rh_inc_pending(ms->rh, &sync);
 757	dm_rh_inc_pending(ms->rh, &nosync);
 758
 759	/*
 760	 * If the flush fails on a previous call and succeeds here,
 761	 * we must not reset the log_failure variable.  We need
 762	 * userspace interaction to do that.
 763	 */
 764	ms->log_failure = dm_rh_flush(ms->rh) ? 1 : ms->log_failure;
 765
 766	/*
 767	 * Dispatch io.
 768	 */
 769	if (unlikely(ms->log_failure) && errors_handled(ms)) {
 770		spin_lock_irq(&ms->lock);
 771		bio_list_merge(&ms->failures, &sync);
 772		spin_unlock_irq(&ms->lock);
 773		wakeup_mirrord(ms);
 774	} else
 775		while ((bio = bio_list_pop(&sync)))
 776			do_write(ms, bio);
 777
 778	while ((bio = bio_list_pop(&recover)))
 779		dm_rh_delay(ms->rh, bio);
 780
 781	while ((bio = bio_list_pop(&nosync))) {
 782		if (unlikely(ms->leg_failure) && errors_handled(ms) && !keep_log(ms)) {
 783			spin_lock_irq(&ms->lock);
 784			bio_list_add(&ms->failures, bio);
 785			spin_unlock_irq(&ms->lock);
 786			wakeup_mirrord(ms);
 787		} else {
 788			map_bio(get_default_mirror(ms), bio);
 789			submit_bio_noacct(bio);
 790		}
 791	}
 792}
 793
 794static void do_failures(struct mirror_set *ms, struct bio_list *failures)
 795{
 796	struct bio *bio;
 797
 798	if (likely(!failures->head))
 799		return;
 800
 801	/*
 802	 * If the log has failed, unattempted writes are being
 803	 * put on the holds list.  We can't issue those writes
 804	 * until a log has been marked, so we must store them.
 805	 *
 806	 * If a 'noflush' suspend is in progress, we can requeue
 807	 * the I/O's to the core.  This give userspace a chance
 808	 * to reconfigure the mirror, at which point the core
 809	 * will reissue the writes.  If the 'noflush' flag is
 810	 * not set, we have no choice but to return errors.
 811	 *
 812	 * Some writes on the failures list may have been
 813	 * submitted before the log failure and represent a
 814	 * failure to write to one of the devices.  It is ok
 815	 * for us to treat them the same and requeue them
 816	 * as well.
 817	 */
 818	while ((bio = bio_list_pop(failures))) {
 819		if (!ms->log_failure) {
 820			ms->in_sync = 0;
 821			dm_rh_mark_nosync(ms->rh, bio);
 822		}
 823
 824		/*
 825		 * If all the legs are dead, fail the I/O.
 826		 * If the device has failed and keep_log is enabled,
 827		 * fail the I/O.
 828		 *
 829		 * If we have been told to handle errors, and keep_log
 830		 * isn't enabled, hold the bio and wait for userspace to
 831		 * deal with the problem.
 832		 *
 833		 * Otherwise pretend that the I/O succeeded. (This would
 834		 * be wrong if the failed leg returned after reboot and
 835		 * got replicated back to the good legs.)
 836		 */
 837		if (unlikely(!get_valid_mirror(ms) || (keep_log(ms) && ms->log_failure)))
 838			bio_io_error(bio);
 839		else if (errors_handled(ms) && !keep_log(ms))
 840			hold_bio(ms, bio);
 841		else
 842			bio_endio(bio);
 843	}
 844}
 845
 846static void trigger_event(struct work_struct *work)
 847{
 848	struct mirror_set *ms =
 849		container_of(work, struct mirror_set, trigger_event);
 850
 851	dm_table_event(ms->ti->table);
 852}
 853
 854/*
 855 *---------------------------------------------------------------
 856 * kmirrord
 857 *---------------------------------------------------------------
 858 */
 859static void do_mirror(struct work_struct *work)
 860{
 861	struct mirror_set *ms = container_of(work, struct mirror_set,
 862					     kmirrord_work);
 863	struct bio_list reads, writes, failures;
 864	unsigned long flags;
 865
 866	spin_lock_irqsave(&ms->lock, flags);
 867	reads = ms->reads;
 868	writes = ms->writes;
 869	failures = ms->failures;
 870	bio_list_init(&ms->reads);
 871	bio_list_init(&ms->writes);
 872	bio_list_init(&ms->failures);
 873	spin_unlock_irqrestore(&ms->lock, flags);
 874
 875	dm_rh_update_states(ms->rh, errors_handled(ms));
 876	do_recovery(ms);
 877	do_reads(ms, &reads);
 878	do_writes(ms, &writes);
 879	do_failures(ms, &failures);
 880}
 881
 882/*
 883 *---------------------------------------------------------------
 884 * Target functions
 885 *---------------------------------------------------------------
 886 */
 887static struct mirror_set *alloc_context(unsigned int nr_mirrors,
 888					uint32_t region_size,
 889					struct dm_target *ti,
 890					struct dm_dirty_log *dl)
 891{
 892	struct mirror_set *ms =
 893		kzalloc(struct_size(ms, mirror, nr_mirrors), GFP_KERNEL);
 894
 895	if (!ms) {
 896		ti->error = "Cannot allocate mirror context";
 897		return NULL;
 898	}
 899
 900	spin_lock_init(&ms->lock);
 901	bio_list_init(&ms->reads);
 902	bio_list_init(&ms->writes);
 903	bio_list_init(&ms->failures);
 904	bio_list_init(&ms->holds);
 905
 906	ms->ti = ti;
 907	ms->nr_mirrors = nr_mirrors;
 908	ms->nr_regions = dm_sector_div_up(ti->len, region_size);
 909	ms->in_sync = 0;
 910	ms->log_failure = 0;
 911	ms->leg_failure = 0;
 912	atomic_set(&ms->suspend, 0);
 913	atomic_set(&ms->default_mirror, DEFAULT_MIRROR);
 914
 915	ms->io_client = dm_io_client_create();
 916	if (IS_ERR(ms->io_client)) {
 917		ti->error = "Error creating dm_io client";
 918		kfree(ms);
 919		return NULL;
 920	}
 921
 922	ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord,
 923				       wakeup_all_recovery_waiters,
 924				       ms->ti->begin, MAX_RECOVERY,
 925				       dl, region_size, ms->nr_regions);
 926	if (IS_ERR(ms->rh)) {
 927		ti->error = "Error creating dirty region hash";
 928		dm_io_client_destroy(ms->io_client);
 929		kfree(ms);
 930		return NULL;
 931	}
 932
 933	return ms;
 934}
 935
 936static void free_context(struct mirror_set *ms, struct dm_target *ti,
 937			 unsigned int m)
 938{
 939	while (m--)
 940		dm_put_device(ti, ms->mirror[m].dev);
 941
 942	dm_io_client_destroy(ms->io_client);
 943	dm_region_hash_destroy(ms->rh);
 944	kfree(ms);
 945}
 946
 947static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
 948		      unsigned int mirror, char **argv)
 949{
 950	unsigned long long offset;
 951	char dummy;
 952	int ret;
 953
 954	if (sscanf(argv[1], "%llu%c", &offset, &dummy) != 1 ||
 955	    offset != (sector_t)offset) {
 956		ti->error = "Invalid offset";
 957		return -EINVAL;
 958	}
 959
 960	ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
 961			    &ms->mirror[mirror].dev);
 962	if (ret) {
 963		ti->error = "Device lookup failure";
 964		return ret;
 965	}
 966
 967	ms->mirror[mirror].ms = ms;
 968	atomic_set(&(ms->mirror[mirror].error_count), 0);
 969	ms->mirror[mirror].error_type = 0;
 970	ms->mirror[mirror].offset = offset;
 971
 972	return 0;
 973}
 974
 975/*
 976 * Create dirty log: log_type #log_params <log_params>
 977 */
 978static struct dm_dirty_log *create_dirty_log(struct dm_target *ti,
 979					     unsigned int argc, char **argv,
 980					     unsigned int *args_used)
 981{
 982	unsigned int param_count;
 983	struct dm_dirty_log *dl;
 984	char dummy;
 985
 986	if (argc < 2) {
 987		ti->error = "Insufficient mirror log arguments";
 988		return NULL;
 989	}
 990
 991	if (sscanf(argv[1], "%u%c", &param_count, &dummy) != 1) {
 992		ti->error = "Invalid mirror log argument count";
 993		return NULL;
 994	}
 995
 996	*args_used = 2 + param_count;
 997
 998	if (argc < *args_used) {
 999		ti->error = "Insufficient mirror log arguments";
1000		return NULL;
1001	}
1002
1003	dl = dm_dirty_log_create(argv[0], ti, mirror_flush, param_count,
1004				 argv + 2);
1005	if (!dl) {
1006		ti->error = "Error creating mirror dirty log";
1007		return NULL;
1008	}
1009
1010	return dl;
1011}
1012
1013static int parse_features(struct mirror_set *ms, unsigned int argc, char **argv,
1014			  unsigned int *args_used)
1015{
1016	unsigned int num_features;
1017	struct dm_target *ti = ms->ti;
1018	char dummy;
1019	int i;
1020
1021	*args_used = 0;
1022
1023	if (!argc)
1024		return 0;
1025
1026	if (sscanf(argv[0], "%u%c", &num_features, &dummy) != 1) {
1027		ti->error = "Invalid number of features";
1028		return -EINVAL;
1029	}
1030
1031	argc--;
1032	argv++;
1033	(*args_used)++;
1034
1035	if (num_features > argc) {
1036		ti->error = "Not enough arguments to support feature count";
1037		return -EINVAL;
1038	}
1039
1040	for (i = 0; i < num_features; i++) {
1041		if (!strcmp("handle_errors", argv[0]))
1042			ms->features |= DM_RAID1_HANDLE_ERRORS;
1043		else if (!strcmp("keep_log", argv[0]))
1044			ms->features |= DM_RAID1_KEEP_LOG;
1045		else {
1046			ti->error = "Unrecognised feature requested";
1047			return -EINVAL;
1048		}
1049
1050		argc--;
1051		argv++;
1052		(*args_used)++;
1053	}
1054	if (!errors_handled(ms) && keep_log(ms)) {
1055		ti->error = "keep_log feature requires the handle_errors feature";
1056		return -EINVAL;
1057	}
1058
1059	return 0;
1060}
1061
1062/*
1063 * Construct a mirror mapping:
1064 *
1065 * log_type #log_params <log_params>
1066 * #mirrors [mirror_path offset]{2,}
1067 * [#features <features>]
1068 *
1069 * log_type is "core" or "disk"
1070 * #log_params is between 1 and 3
1071 *
1072 * If present, supported features are "handle_errors" and "keep_log".
1073 */
1074static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1075{
1076	int r;
1077	unsigned int nr_mirrors, m, args_used;
1078	struct mirror_set *ms;
1079	struct dm_dirty_log *dl;
1080	char dummy;
1081
1082	dl = create_dirty_log(ti, argc, argv, &args_used);
1083	if (!dl)
1084		return -EINVAL;
1085
1086	argv += args_used;
1087	argc -= args_used;
1088
1089	if (!argc || sscanf(argv[0], "%u%c", &nr_mirrors, &dummy) != 1 ||
1090	    nr_mirrors < 2 || nr_mirrors > MAX_NR_MIRRORS) {
1091		ti->error = "Invalid number of mirrors";
1092		dm_dirty_log_destroy(dl);
1093		return -EINVAL;
1094	}
1095
1096	argv++, argc--;
1097
1098	if (argc < nr_mirrors * 2) {
1099		ti->error = "Too few mirror arguments";
1100		dm_dirty_log_destroy(dl);
1101		return -EINVAL;
1102	}
1103
1104	ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
1105	if (!ms) {
1106		dm_dirty_log_destroy(dl);
1107		return -ENOMEM;
1108	}
1109
1110	/* Get the mirror parameter sets */
1111	for (m = 0; m < nr_mirrors; m++) {
1112		r = get_mirror(ms, ti, m, argv);
1113		if (r) {
1114			free_context(ms, ti, m);
1115			return r;
1116		}
1117		argv += 2;
1118		argc -= 2;
1119	}
1120
1121	ti->private = ms;
1122
1123	r = dm_set_target_max_io_len(ti, dm_rh_get_region_size(ms->rh));
1124	if (r)
1125		goto err_free_context;
1126
1127	ti->num_flush_bios = 1;
1128	ti->num_discard_bios = 1;
1129	ti->per_io_data_size = sizeof(struct dm_raid1_bio_record);
1130
1131	ms->kmirrord_wq = alloc_workqueue("kmirrord", WQ_MEM_RECLAIM, 0);
1132	if (!ms->kmirrord_wq) {
1133		DMERR("couldn't start kmirrord");
1134		r = -ENOMEM;
1135		goto err_free_context;
1136	}
1137	INIT_WORK(&ms->kmirrord_work, do_mirror);
1138	timer_setup(&ms->timer, delayed_wake_fn, 0);
1139	ms->timer_pending = 0;
1140	INIT_WORK(&ms->trigger_event, trigger_event);
1141
1142	r = parse_features(ms, argc, argv, &args_used);
1143	if (r)
1144		goto err_destroy_wq;
1145
1146	argv += args_used;
1147	argc -= args_used;
1148
1149	/*
1150	 * Any read-balancing addition depends on the
1151	 * DM_RAID1_HANDLE_ERRORS flag being present.
1152	 * This is because the decision to balance depends
1153	 * on the sync state of a region.  If the above
1154	 * flag is not present, we ignore errors; and
1155	 * the sync state may be inaccurate.
1156	 */
1157
1158	if (argc) {
1159		ti->error = "Too many mirror arguments";
1160		r = -EINVAL;
1161		goto err_destroy_wq;
1162	}
1163
1164	ms->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1165	if (IS_ERR(ms->kcopyd_client)) {
1166		r = PTR_ERR(ms->kcopyd_client);
1167		goto err_destroy_wq;
1168	}
1169
1170	wakeup_mirrord(ms);
1171	return 0;
1172
1173err_destroy_wq:
1174	destroy_workqueue(ms->kmirrord_wq);
1175err_free_context:
1176	free_context(ms, ti, ms->nr_mirrors);
1177	return r;
1178}
1179
1180static void mirror_dtr(struct dm_target *ti)
1181{
1182	struct mirror_set *ms = ti->private;
1183
1184	timer_delete_sync(&ms->timer);
1185	flush_workqueue(ms->kmirrord_wq);
1186	flush_work(&ms->trigger_event);
1187	dm_kcopyd_client_destroy(ms->kcopyd_client);
1188	destroy_workqueue(ms->kmirrord_wq);
1189	free_context(ms, ti, ms->nr_mirrors);
1190}
1191
1192/*
1193 * Mirror mapping function
1194 */
1195static int mirror_map(struct dm_target *ti, struct bio *bio)
1196{
1197	int r, rw = bio_data_dir(bio);
1198	struct mirror *m;
1199	struct mirror_set *ms = ti->private;
1200	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1201	struct dm_raid1_bio_record *bio_record =
1202	  dm_per_bio_data(bio, sizeof(struct dm_raid1_bio_record));
1203
1204	bio_record->details.bi_bdev = NULL;
1205
1206	if (rw == WRITE) {
1207		/* Save region for mirror_end_io() handler */
1208		bio_record->write_region = dm_rh_bio_to_region(ms->rh, bio);
1209		queue_bio(ms, bio, rw);
1210		return DM_MAPIO_SUBMITTED;
1211	}
1212
1213	r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0);
1214	if (r < 0 && r != -EWOULDBLOCK)
1215		return DM_MAPIO_KILL;
1216
1217	/*
1218	 * If region is not in-sync queue the bio.
1219	 */
1220	if (!r || (r == -EWOULDBLOCK)) {
1221		if (bio->bi_opf & REQ_RAHEAD)
1222			return DM_MAPIO_KILL;
1223
1224		queue_bio(ms, bio, rw);
1225		return DM_MAPIO_SUBMITTED;
1226	}
1227
1228	/*
1229	 * The region is in-sync and we can perform reads directly.
1230	 * Store enough information so we can retry if it fails.
1231	 */
1232	m = choose_mirror(ms, bio->bi_iter.bi_sector);
1233	if (unlikely(!m))
1234		return DM_MAPIO_KILL;
1235
1236	dm_bio_record(&bio_record->details, bio);
1237	bio_record->m = m;
1238
1239	map_bio(m, bio);
1240
1241	return DM_MAPIO_REMAPPED;
1242}
1243
1244static int mirror_end_io(struct dm_target *ti, struct bio *bio,
1245		blk_status_t *error)
1246{
1247	int rw = bio_data_dir(bio);
1248	struct mirror_set *ms = ti->private;
1249	struct mirror *m = NULL;
1250	struct dm_bio_details *bd = NULL;
1251	struct dm_raid1_bio_record *bio_record =
1252	  dm_per_bio_data(bio, sizeof(struct dm_raid1_bio_record));
1253
1254	/*
1255	 * We need to dec pending if this was a write.
1256	 */
1257	if (rw == WRITE) {
1258		if (!(bio->bi_opf & REQ_PREFLUSH) &&
1259		    bio_op(bio) != REQ_OP_DISCARD)
1260			dm_rh_dec(ms->rh, bio_record->write_region);
1261		return DM_ENDIO_DONE;
1262	}
1263
1264	if (*error == BLK_STS_NOTSUPP)
1265		goto out;
1266
1267	if (bio->bi_opf & REQ_RAHEAD)
1268		goto out;
1269
1270	if (unlikely(*error)) {
1271		if (!bio_record->details.bi_bdev) {
1272			/*
1273			 * There wasn't enough memory to record necessary
1274			 * information for a retry or there was no other
1275			 * mirror in-sync.
1276			 */
1277			DMERR_LIMIT("Mirror read failed.");
1278			return DM_ENDIO_DONE;
1279		}
1280
1281		m = bio_record->m;
1282
1283		DMERR("Mirror read failed from %s. Trying alternative device.",
1284		      m->dev->name);
1285
1286		fail_mirror(m, DM_RAID1_READ_ERROR);
1287
1288		/*
1289		 * A failed read is requeued for another attempt using an intact
1290		 * mirror.
1291		 */
1292		if (default_ok(m) || mirror_available(ms, bio)) {
1293			bd = &bio_record->details;
1294
1295			dm_bio_restore(bd, bio);
1296			bio_record->details.bi_bdev = NULL;
1297			bio->bi_status = 0;
1298
1299			queue_bio(ms, bio, rw);
1300			return DM_ENDIO_INCOMPLETE;
1301		}
1302		DMERR("All replicated volumes dead, failing I/O");
1303	}
1304
1305out:
1306	bio_record->details.bi_bdev = NULL;
1307
1308	return DM_ENDIO_DONE;
1309}
1310
1311static void mirror_presuspend(struct dm_target *ti)
1312{
1313	struct mirror_set *ms = ti->private;
1314	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1315
1316	struct bio_list holds;
1317	struct bio *bio;
1318
1319	atomic_set(&ms->suspend, 1);
1320
1321	/*
1322	 * Process bios in the hold list to start recovery waiting
1323	 * for bios in the hold list. After the process, no bio has
1324	 * a chance to be added in the hold list because ms->suspend
1325	 * is set.
1326	 */
1327	spin_lock_irq(&ms->lock);
1328	holds = ms->holds;
1329	bio_list_init(&ms->holds);
1330	spin_unlock_irq(&ms->lock);
1331
1332	while ((bio = bio_list_pop(&holds)))
1333		hold_bio(ms, bio);
1334
1335	/*
1336	 * We must finish up all the work that we've
1337	 * generated (i.e. recovery work).
1338	 */
1339	dm_rh_stop_recovery(ms->rh);
1340
1341	wait_event(_kmirrord_recovery_stopped,
1342		   !dm_rh_recovery_in_flight(ms->rh));
1343
1344	if (log->type->presuspend && log->type->presuspend(log))
1345		/* FIXME: need better error handling */
1346		DMWARN("log presuspend failed");
1347
1348	/*
1349	 * Now that recovery is complete/stopped and the
1350	 * delayed bios are queued, we need to wait for
1351	 * the worker thread to complete.  This way,
1352	 * we know that all of our I/O has been pushed.
1353	 */
1354	flush_workqueue(ms->kmirrord_wq);
1355}
1356
1357static void mirror_postsuspend(struct dm_target *ti)
1358{
1359	struct mirror_set *ms = ti->private;
1360	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1361
1362	if (log->type->postsuspend && log->type->postsuspend(log))
1363		/* FIXME: need better error handling */
1364		DMWARN("log postsuspend failed");
1365}
1366
1367static void mirror_resume(struct dm_target *ti)
1368{
1369	struct mirror_set *ms = ti->private;
1370	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1371
1372	atomic_set(&ms->suspend, 0);
1373	if (log->type->resume && log->type->resume(log))
1374		/* FIXME: need better error handling */
1375		DMWARN("log resume failed");
1376	dm_rh_start_recovery(ms->rh);
1377}
1378
1379/*
1380 * device_status_char
1381 * @m: mirror device/leg we want the status of
1382 *
1383 * We return one character representing the most severe error
1384 * we have encountered.
1385 *    A => Alive - No failures
1386 *    D => Dead - A write failure occurred leaving mirror out-of-sync
1387 *    S => Sync - A sychronization failure occurred, mirror out-of-sync
1388 *    R => Read - A read failure occurred, mirror data unaffected
1389 *
1390 * Returns: <char>
1391 */
1392static char device_status_char(struct mirror *m)
1393{
1394	if (!atomic_read(&(m->error_count)))
1395		return 'A';
1396
1397	return (test_bit(DM_RAID1_FLUSH_ERROR, &(m->error_type))) ? 'F' :
1398		(test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' :
1399		(test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' :
1400		(test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U';
1401}
1402
1403
1404static void mirror_status(struct dm_target *ti, status_type_t type,
1405			  unsigned int status_flags, char *result, unsigned int maxlen)
1406{
1407	unsigned int m, sz = 0;
1408	int num_feature_args = 0;
1409	struct mirror_set *ms = ti->private;
1410	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1411	char buffer[MAX_NR_MIRRORS + 1];
1412
1413	switch (type) {
1414	case STATUSTYPE_INFO:
1415		DMEMIT("%d ", ms->nr_mirrors);
1416		for (m = 0; m < ms->nr_mirrors; m++) {
1417			DMEMIT("%s ", ms->mirror[m].dev->name);
1418			buffer[m] = device_status_char(&(ms->mirror[m]));
1419		}
1420		buffer[m] = '\0';
1421
1422		DMEMIT("%llu/%llu 1 %s ",
1423		      (unsigned long long)log->type->get_sync_count(log),
1424		      (unsigned long long)ms->nr_regions, buffer);
1425
1426		sz += log->type->status(log, type, result+sz, maxlen-sz);
1427
1428		break;
1429
1430	case STATUSTYPE_TABLE:
1431		sz = log->type->status(log, type, result, maxlen);
1432
1433		DMEMIT("%d", ms->nr_mirrors);
1434		for (m = 0; m < ms->nr_mirrors; m++)
1435			DMEMIT(" %s %llu", ms->mirror[m].dev->name,
1436			       (unsigned long long)ms->mirror[m].offset);
1437
1438		num_feature_args += !!errors_handled(ms);
1439		num_feature_args += !!keep_log(ms);
1440		if (num_feature_args) {
1441			DMEMIT(" %d", num_feature_args);
1442			if (errors_handled(ms))
1443				DMEMIT(" handle_errors");
1444			if (keep_log(ms))
1445				DMEMIT(" keep_log");
1446		}
1447
1448		break;
1449
1450	case STATUSTYPE_IMA:
1451		DMEMIT_TARGET_NAME_VERSION(ti->type);
1452		DMEMIT(",nr_mirrors=%d", ms->nr_mirrors);
1453		for (m = 0; m < ms->nr_mirrors; m++) {
1454			DMEMIT(",mirror_device_%d=%s", m, ms->mirror[m].dev->name);
1455			DMEMIT(",mirror_device_%d_status=%c",
1456			       m, device_status_char(&(ms->mirror[m])));
1457		}
1458
1459		DMEMIT(",handle_errors=%c", errors_handled(ms) ? 'y' : 'n');
1460		DMEMIT(",keep_log=%c", keep_log(ms) ? 'y' : 'n');
1461
1462		DMEMIT(",log_type_status=");
1463		sz += log->type->status(log, type, result+sz, maxlen-sz);
1464		DMEMIT(";");
1465		break;
1466	}
1467}
1468
1469static int mirror_iterate_devices(struct dm_target *ti,
1470				  iterate_devices_callout_fn fn, void *data)
1471{
1472	struct mirror_set *ms = ti->private;
1473	int ret = 0;
1474	unsigned int i;
1475
1476	for (i = 0; !ret && i < ms->nr_mirrors; i++)
1477		ret = fn(ti, ms->mirror[i].dev,
1478			 ms->mirror[i].offset, ti->len, data);
1479
1480	return ret;
1481}
1482
1483static struct target_type mirror_target = {
1484	.name	 = "mirror",
1485	.version = {1, 15, 0},
1486	.module	 = THIS_MODULE,
1487	.features = DM_TARGET_ATOMIC_WRITES,
1488	.ctr	 = mirror_ctr,
1489	.dtr	 = mirror_dtr,
1490	.map	 = mirror_map,
1491	.end_io	 = mirror_end_io,
1492	.presuspend = mirror_presuspend,
1493	.postsuspend = mirror_postsuspend,
1494	.resume	 = mirror_resume,
1495	.status	 = mirror_status,
1496	.iterate_devices = mirror_iterate_devices,
1497};
1498
1499static int __init dm_mirror_init(void)
1500{
1501	int r;
1502
1503	dm_raid1_wq = alloc_workqueue("dm_raid1_wq", 0, 0);
1504	if (!dm_raid1_wq) {
1505		DMERR("Failed to alloc workqueue");
1506		return -ENOMEM;
1507	}
1508
1509	r = dm_register_target(&mirror_target);
1510	if (r < 0) {
1511		destroy_workqueue(dm_raid1_wq);
1512		return r;
1513	}
1514
1515	return 0;
1516}
1517
1518static void __exit dm_mirror_exit(void)
1519{
1520	destroy_workqueue(dm_raid1_wq);
1521	dm_unregister_target(&mirror_target);
1522}
1523
1524/* Module hooks */
1525module_init(dm_mirror_init);
1526module_exit(dm_mirror_exit);
1527
1528MODULE_DESCRIPTION(DM_NAME " mirror target");
1529MODULE_AUTHOR("Joe Thornber");
1530MODULE_LICENSE("GPL");